[Dovecot] Best location for indexes?
I cant seem to find anything via a google search, so if this has been answered I apologize.
I am wondering if INDEX=MEMORY is going to give the best performance versus files? I have a server with directly attached storage, 8 gigs of ram, about 500 gigs of data.
I am ok with losing and having to rebuild the indexes on a reboot as the last uptime was over 200 days
On 05/11/10 16:05, Brandon Lamb wrote:
I cant seem to find anything via a google search, so if this has been answered I apologize.
I am wondering if INDEX=MEMORY is going to give the best performance versus files? I have a server with directly attached storage, 8 gigs of ram, about 500 gigs of data.
I am ok with losing and having to rebuild the indexes on a reboot as the last uptime was over 200 days
Last I checked, INDEX=MEMORY rebuilds the index on demand, on each _connection_.
Your best bet is to keep the indices on disk. Certainly, if you can keep them on a separate spindle from the mail store, you may find a performance improvement through parallelism and lack of seek clash.
-- Curtis Maloney
On Mon, May 10, 2010 at 11:21 PM, Curtis Maloney cmaloney@cardgate.net wrote:
On 05/11/10 16:05, Brandon Lamb wrote:
I cant seem to find anything via a google search, so if this has been answered I apologize.
I am wondering if INDEX=MEMORY is going to give the best performance versus files? I have a server with directly attached storage, 8 gigs of ram, about 500 gigs of data.
I am ok with losing and having to rebuild the indexes on a reboot as the last uptime was over 200 days
Last I checked, INDEX=MEMORY rebuilds the index on demand, on each _connection_.
Your best bet is to keep the indices on disk. Certainly, if you can keep them on a separate spindle from the mail store, you may find a performance improvement through parallelism and lack of seek clash.
-- Curtis Maloney
Oh damnit haha. I thought it saved it in memory like a ramdisk or some kind of implementation like that, not only the duration of a connection =\
Back to disk then!
Thanks! =)
On 11/05/10 08:29, Brandon Lamb wrote:
On Mon, May 10, 2010 at 11:21 PM, Curtis Maloney cmaloney@cardgate.net wrote:
On 05/11/10 16:05, Brandon Lamb wrote:
I cant seem to find anything via a google search, so if this has been answered I apologize.
I am wondering if INDEX=MEMORY is going to give the best performance versus files? I have a server with directly attached storage, 8 gigs of ram, about 500 gigs of data.
I am ok with losing and having to rebuild the indexes on a reboot as the last uptime was over 200 days
Last I checked, INDEX=MEMORY rebuilds the index on demand, on each _connection_.
Your best bet is to keep the indices on disk. Certainly, if you can keep them on a separate spindle from the mail store, you may find a performance improvement through parallelism and lack of seek clash.
-- Curtis Maloney
Oh damnit haha. I thought it saved it in memory like a ramdisk or some kind of implementation like that, not only the duration of a connection =\
Back to disk then!
Thanks! =)
You could implement your own suggestion easily with a disk implementation, being 'the disk' a tmpfs or similar partition.
Regards, Tom
participants (3)
-
Brandon Lamb
-
Curtis Maloney
-
Tom Hendrikx